"We can't measure ROI on digital marketing." It's one of the sentences I hear most often. And it's usually not true — but it's understandable. Digital attribution is complex, the customer journey is rarely linear, and data quality is never perfect. It's tempting to conclude that precise ROI measurement is an illusion.
But that's the wrong conclusion. Precise ROI measurement is an illusion. Useful ROI measurement is not — and that's what we should aim for.
Here's the methodological approach I use with companies working in complex, multi-channel setups with imperfect data.
Phase 1: Define what you actually want to move
It sounds banal, but it's fundamental. Most companies have too many metrics. They track 30 to 60 KPIs across channels and platforms — and none of them is clearly linked to business outcomes. Result: everyone can see the numbers moving, but no one can say whether the company is actually making more money.
Start with one business metric that matters. Not "we want more traffic" or "we want to improve our ROAS". Define the business end-metric: revenue from digitally acquired customers, number of new B2B customers with revenue above X, or average lifetime value of customers acquired via specific channels. Everything else is a proxy metric — useful for optimisation, but not the ultimate goal.
Phase 2: Build an incrementality framework rather than an attribution framework
The classic approach to digital ROI is attribution modelling: assign credit to the channels and touchpoints that contributed to a conversion. It's a useful perspective — but it's fundamentally the wrong question. Attribution asks: who should get credit? Incrementality asks: what would have happened if we hadn't run this activity?
These two questions give very different answers — and it's incrementality that tells you the actual ROI. A channel can have high attribution and low incrementality: you give it credit for customers who would have bought anyway. Conversely, a channel can have low attribution and high incrementality: it actually creates new customers, but they convert via another channel, thereby not giving credit to the channel that actually started the journey.
The practical approach to incrementality measurement is geo-tests and holdout groups. Run a campaign in one geographic market and hold a comparable market without the campaign. Compare revenue development in the two markets. The difference is the incremental effect of the campaign. It's not perfect — geographic markets are never identical — but it's far more credible than attribution modelling alone.
Phase 3: Use Marketing Mix Modelling as a strategic compass
Attribution data and incrementality tests give you a good picture of what's happening tactically. But they don't tell you the strategic picture: What is the relative contribution from your different channels over time, and what happens when you shift budget allocation?
Marketing Mix Modelling (MMM) addresses this. MMM is a statistical technique that estimates channel contribution based on historical data and statistical correlations. It handles the fact that marketing effects aren't instantaneous (there are lag effects: a campaign today can drive revenue three months from now), and that channels interact with each other (TV advertising can amplify the effect of digital performance marketing).
MMM requires solid historical data — typically at least two years of weekly data on revenue, channel spend and exogenous factors such as seasonality and competitor activity. It's not easy to implement. But for companies with complex, multi-channel setups and revenue above 50-100 million DKK, it's the most reliable strategic guide for budget allocation.
Phase 4: Define and communicate the uncertainty
This is where most ROI reports fail. They present numbers with false precision: "Our paid social channel generated 3.4x ROAS." But what is the uncertainty on that number? Is it 3.4x ± 0.2 or ± 1.5? That's a critical difference for the decisions the numbers are meant to support.
Be explicit about the uncertainty in your ROI estimates. Present it as intervals, not point estimates: "We estimate that the paid social channel delivers ROAS in the range 2.8-4.0, with greatest probability around 3.4." That's more honest and actually more credible than a precise single number that gives a false sense of certainty.
Phase 5: Create a reporting cadence that matches the decision cycle
This is the most overlooked element in ROI measurement. Data is only useful if it's available when decisions are made. If your quarterly budget meeting is on the 15th of February, and your quarterly analytics report is ready on the 20th, the reporting is useless for that meeting.
Map your decision cycle. When are the big budget decisions made? When are campaigns optimised? When is channel strategy evaluated? Design your reporting cadence and data availability around these decision points — not around what's technically easy to generate.
ROI measurement is not a technical problem. It's a methodological and communicative problem. The method is about asking the right questions — incrementality over attribution, intervals over point estimates, business outcomes over proxy metrics. The communication is about presenting it to the right people at the right times with the right uncertainty calibration.
Companies that master both make markedly better marketing decisions — not because they have better data, but because they use the data they have in a more intelligent way.